mohammed/whisper-small-arabic-202505

Whisper Small AR - Mohammed Bakheet

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2732
• Wer: 21.5262

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.2079	250	0.3651	29.8066
0.5126	0.4158	500	0.3310	27.5784
0.5126	0.6237	750	0.3087	25.3032
0.2513	0.8316	1000	0.2865	24.4490
0.2513	1.0399	1250	0.2761	23.2251
0.1679	1.2478	1500	0.2755	22.9491
0.1679	1.4557	1750	0.2692	22.4329
0.1343	1.6636	2000	0.2682	22.0086
0.1343	1.8715	2250	0.2629	21.6670
0.1159	2.0798	2500	0.2669	21.5600
0.1159	2.2877	2750	0.2732	21.5262

Framework versions

• Transformers 4.51.3
• Pytorch 2.5.1+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1